Bug 32654 (Slow, startup)

If that message shows up before the delay, that's unlikely to be the cause.
Whatever comes next is most likely responsible.

What's the output of "systemd-analyze blame|head"?

If the delay is not shown there, run "journalctl -b --no-h>jc.txt"
and attach the file to this bug report.

CC: (none) => davidwhodgins

We have a bug with amdgpupro-opencl-orca bug#32649

Hi katnatek.

The bug in reference prevents to download the OpenCL proprietary driver.
It means nothing is installed by it. Moreover, amdgpupro-opencl-orca brings OpenCL computational capabilities. So, I doubt the cause is here. I would follow Dave's path.

Video rendering is more managed by x11-driver-video-amdgpu.

As far as OpenCL is concerned, amdgpupro-opencl-orca is suitable for older generation but, potentially, not as old as this Radeon HD 8530M.
Not being able to be exhaustive to cover all models, a GPU architecture more recent than GCN 2nd generation might be required; the GPU might need to support OpenCL 2.0, at least.
The examples I gave on amdgpupro-opencl-orca description (Grenada XT, Carrizo and Polaris) meet these requirements.
I cannot be more precise because there is no official documentation from AMD for this driver. At least, I haven't found any...

If you can now try to install amdgpupro-opencl-orca from nonfree/updates_testing, we'll not whether the above requirements are strict or whether even older GPU can be supported. 

Thanks.

(In reply to christian squidf from comment #3)
I add you to CC and link to the bug because reporter mention the package
I thank you the information, now we need reporter feedback to Dave's suggestion

(In reply to Dave Hodgins from comment #1)
> If that message shows up before the delay, that's unlikely to be the cause.
> Whatever comes next is most likely responsible.
> 
> What's the output of "systemd-analyze blame|head"?
> 
> If the delay is not shown there, run "journalctl -b --no-h>jc.txt"
> and attach the file to this bug report.

You are probably right: there are two problems reported under this bug heading.

systemd-analyze blame|head
2min 201ms systemd-networkd-wait-online.service
    6.285s network.service
    3.605s docker.service
    2.678s mandriva-everytime.service
    1.835s postfix.service
    1.494s systemd-udev-settle.service
    1.431s shorewall6.service
    1.190s blueman-mechanism.service
     997ms udisks2.service
     956ms accounts-daemon.service

So, the delay culprit seems to be network-wait-on-line.service. However, when I disabled it on booting the first row of systemd-analyze disappears, but the 2 min waiting is still there.

Created attachment 14236 [details]
output of journalctl -b --no-h>jc.txt

Completing the reply to Dave

Decide how you want to manage the network connections. In that journal output
it shows ...

systemd-networkd[3734]: wlp3s0: Connected WiFi access point: NETGEAR88_EXT2 (bc:a5:11:1d:54:9a)
NetworkManager[4065]: <info>  [1703751570.2702] dhcp4 (wlp3s0): state changed new lease, address=192.168.0.23
network[4592]: Error: Connection activation failed: The base network connection was interrupted

While I've used systemd-networkd and the network scripts to get ipv6 working
properly on my connection in the past, it's fragile.

Use one of systemd-networkd, NetworkManager, or network. Not all three as
they are interfering with each other.

I recommend NetworkManager.
See https://wiki.mageia.org/en/Switching_to_networkmanager
It doesn't mention systemd-networkd as it's disabled by default. If switching
to only using NetworkManager, disable systemd-networkd again.

Also, change ONBOOT=yes to ONBOOT=no in all of the files shown in
/etc/sysconfig/network-scripts/ifcfg*
that are for devices not currently present, or not currently being used.

(In reply to Alberto Girlando from comment #0)
> Old ASUS portable with two videocards: intel integrated and AMD Radeon HD
> 8530M
[..]
> c) Just at the beginning of startup, after the message about ACPI, systemd
> issues the message: "kdf: amdgpu: OLAND not supported in kfd" then stops for
> about two minutes trying to rescue. But it fails and proceed.

This message is harmless, according to what I read, and you can make it disappear with the option "vce=0".
It seems not related to your problem, anyway.

Is your problem fixed using Dave suggestions ?

CC: (none) => geex+mageia

Thank you all for the help. The problem has been solved following Dave suggestions, although the problem was neither amdgpu (from this point of view, my title was misleading, I also put an alias title, more generic), nor the network managers.

Dave suggestions about the conflicts between the three network managers, was very helpful, and helped me to clean up networking aspect in the old computer as well as in my main one, where I installed Mageia9 from scratch. Indeed I hope next Mageia release abandon completely the old system, always gave problems to me.

As I said, clearing up small delays connected with the network managers did not solve the problem of the 2 minutes delay, only speed up the subsequent steps. But Dave suggestion about the fact that the problem was occurring after the OLAND message, and a close inspection of the report of "journalctl -b --no-h" lead me to consider a warning by dracut about a device not found. And an inspection of dracut.conf and also grub helped me to find the culprit: a wrong labeling of a "resume" optional disk, remained there from an old installation (Mageia 7, perhaps). Sorry signaling this as a bug, it really was MY bug !

To reply to Christian: I tried to install amdgpupro-opencl-orca from nonfree/updates_testing, but apparently did not work. But a previous attempt from the normal repository apparently did something like a fast downloading from somewhere, so maybe the file was already on my computer (how can I check or remove it ?).  Anyway, also using x11-driver-video-amdgpu did not work for my AMD video card. A final note: on the AMD site there is the possibility to have proprietary drivers which apparently include my card. But the rpm file, which simply add the AMD repository, works only for REDHAT distributions (I tried to cheat, but did not work).

Thanks again

Status: NEW => RESOLVED
Resolution: (none) => FIXED
Alias: (none) => Slow, startup

(In reply to Alberto Girlando from comment #9)
> it really was MY bug !

I'd rather say we miss some sanity checks when closing the lid / suspending /hibernating.

(In reply to Alberto Girlando from comment #9)
> To reply to Christian: I tried to install amdgpupro-opencl-orca from
> nonfree/updates_testing, but apparently did not work. But a previous attempt
> from the normal repository apparently did something like a fast downloading
> from somewhere, so maybe the file was already on my computer (how can I
> check or remove it ?). 
> 
Hi. You can look into /tmp/amdgpupro-opencl-orca but installing the package means downloading; there is no test of previous download as it might have disappeared.

Could you please try to install amdgpupro-opencl-orca from nonfree/updates_testing in a console (su -c "urpmi amdgpupro-opencl-orca", after having enabled nonfree/updates_testing in the MCC, for instance).
You will still whether the download has happened properly.

Please, do note this package will not act on your display or primer video rendering. This package is to bring OpenCL capability *only*; it is more to speed up computational speed. It is a small part of the amdgpupro driver.

I don't think this package will do anything in perspective of the issue you are facing; assuming I have well understood it.

(In reply to christian squidf from comment #11)
> (In reply to Alberto Girlando from comment #9)
> > To reply to Christian: I tried to install amdgpupro-opencl-orca from
> > nonfree/updates_testing, but apparently did not work. But a previous attempt
> > from the normal repository apparently did something like a fast downloading
> > from somewhere, so maybe the file was already on my computer (how can I
> > check or remove it ?). 
> > 
> Hi. You can look into /tmp/amdgpupro-opencl-orca but installing the package
> means downloading; there is no test of previous download as it might have
> disappeared.
> 
> Could you please try to install amdgpupro-opencl-orca from
> nonfree/updates_testing in a console (su -c "urpmi amdgpupro-opencl-orca",
> after having enabled nonfree/updates_testing in the MCC, for instance).
> You will still whether the download has happened properly.
> 
> Please, do note this package will not act on your display or primer video
> rendering. This package is to bring OpenCL capability *only*; it is more to
> speed up computational speed. It is a small part of the amdgpupro driver.
> 
> I don't think this package will do anything in perspective of the issue you
> are facing; assuming I have well understood it.


OK, I tried to install amdgpu-opencl-orca from console, and it worked.
I (now) understand that should make my old AMD Radeon HD 8530M work. And no, also AMD site seems not providing driver anymore, except this old 2015 version (Catalyst). I will contiue to use the only intel embedded graphics.

A proprietary driver from 2014/2015 which only works with very old kernels and XFree86 or X.org versions won't get you anywhere.

But there is still hope in the open-source world, your GCN1 card may be supported by radeon or radeonhd module.